christian ruff laboratory for social and neural systems research department of economics university...

Christian Ruff

Laboratory for Social and Neural Systems ResearchDepartment of Economics

University of Zurich

Institute of NeurologyUniversity College London

With thanks to the FIL methods group, in particular Rik Henson

Experimental designExperimental designExperimental designExperimental design

RealignmentRealignment SmoothingSmoothing

NormalisationNormalisation

General linear modelGeneral linear model

Statistical parametric map (SPM)Statistical parametric map (SPM)Image time-seriesImage time-series

Parameter estimatesParameter estimates

Design matrixDesign matrix

TemplateTemplate

KernelKernel

Gaussian Gaussian field theoryfield theory

p <0.05p <0.05

StatisticalStatisticalinferenceinference

• Categorical designsSubtraction - pure insertion, baseline problems

Conjunction - testing multiple hypotheses

• Parametric designsLinear - adaptation, cognitive dimensions

Nonlinear - polynomial expansion

Model-based - algorithmic definition

• Factorial designsCategorical - Main effects and Interactions

Parametric - Psychophysiological interactions

OverviewOverviewOverviewOverview

Subtraction designsSubtraction designsSubtraction designsSubtraction designs

• Question:

– Brain activity supporting process P?

• Procedure:

– Contrast: [Task with P] – [control task without P ] = P

• The critical assumption of “pure insertion”:

– Cognitive (and neural) processes can be added to others without changing them

– Changed behavior (and brain activity) reflects only added process

Subtraction designsSubtraction designsSubtraction designsSubtraction designs

• Example: Brain activity involved in face recognition?

• Question:

– Brain activity supporting process P?

• Procedure:

– Contrast: [Task with P] – [control task without P ] = P

– = face recognition (?)

-- Several components differ ! Several components differ !

• „ „Distant“ stimuli Distant“ stimuli

Subtraction designs: Baseline problemsSubtraction designs: Baseline problems

- Implicit processes in control condition ?

„ „Kate“ „Kate? Or my friend Josie?”Kate“ „Kate? Or my friend Josie?”

• „Related“ stimuli

Name Person! Name Gender!Name Person! Name Gender!

-- Interaction of process and task Interaction of process and task ??

• Same stimuli, different taskSame stimuli, different task

Inferotemporal response to facesSPM{F} testing for

evoked responses

• “ “Baseline” here corresponds to session mean, and thus processing during “rest”Baseline” here corresponds to session mean, and thus processing during “rest”

• Problems:Problems:

- Null events or long SOAs essential for estimationNull events or long SOAs essential for estimation

- ““Cognitive” interpretation hardly possible, as cognitive processes during rest unclear, but Cognitive” interpretation hardly possible, as cognitive processes during rest unclear, but useful to define regions involved in taskuseful to define regions involved in task

Rest baseline: Evoked responsesRest baseline: Evoked responses

Example I: Face selectivity in fusiform Example I: Face selectivity in fusiform gyrusgyrus

Example I: Face selectivity in fusiform Example I: Face selectivity in fusiform gyrusgyrus

Kanwisher, McDermott, Chun (1997) JNeurosci

Example II: Repetition suppressionExample II: Repetition suppressionExample II: Repetition suppressionExample II: Repetition suppression

Henson, Dolan, Shallice (2000) Science

Henson et al (2002) Cereb Cortex

– Repeated viewing of the same face elicits lower BOLD activity in face-selective regions

– Repetition suppression / adaptation designs: BOLD decreases for repetition used to infer functional specialization for this task/stimulus

Example III: MVPAExample III: MVPAExample III: MVPAExample III: MVPA

Haynes & Rees (2004) Nat Neurosci

Kamitani & Tong (2004) Nat Neurosci

– Focus on multivariate activity patterns across voxels (rather than univariate signal in each voxel)

– Many MVPA designs involve categorical comparisons between 2 types of stimuli (and hence all associated baseline/pure insertion problems)

• Categorical designsSubtraction - pure insertion, baseline problems

Conjunction - testing multiple hypotheses

• Parametric designsLinear - adaptation, cognitive dimensions

Nonlinear - polynomial expansion

Model-based - algorithmic definition

• Factorial designsCategorical - Main effects and Interactions

Parametric - Linear and nonlinear interactions

OverviewOverviewOverviewOverview

• One way to minimise the baseline/pure insertion problem is to isolate the same process by two or more separate comparisons, and inspect the resulting simple effects for commonalities

• A test for such activation common to several independent contrasts is called “Conjunction”

ConjunctionsConjunctionsConjunctionsConjunctions

• Conjunctions can be conducted across a whole variety of different contexts:- Tasks- Stimuli- Senses (vision, audition)- etc.

• But the contrasts entering a conjunction have to be truly independent!

Example: Example:

Which neural structures support object recognition, Which neural structures support object recognition,

independent of task (naming vs viewing)?independent of task (naming vs viewing)?

ConjunctionsConjunctionsConjunctionsConjunctions

A1A1 A2A2

B2B2B1B1

Task (1/2)Task (1/2)

Stim

uli (

A/B

)

Col

ours

O

bjec

ts

Naming ViewingNaming Viewing

Price & Friston (1997) Neuroimage

Example: Example:

Which neural structures support object recognition, Which neural structures support object recognition,

independent of task (naming vs viewing)?independent of task (naming vs viewing)?

ConjunctionsConjunctionsConjunctionsConjunctions

A1A1 A2A2

B2B2B1B1

Task (1/2)Task (1/2)

Stim

uli (

A/B

)

Col

ours

O

bjec

ts

Naming ViewingNaming Viewing

Visual Processing (Vis)Visual Processing (Vis)

Object Recognition (Rec)Object Recognition (Rec)

Word Retrieval (Word)Word Retrieval (Word)

VisVisRecRec

VisVisRecRec

WordWord

VisVisVisVisWordWord

&& ==(Rec) (Rec) (Vis)(Vis)

(Rec) (Rec)

(Word)(Word)

(Vis) (Vis)

(Word)(Word)--

Object Object

Naming Naming Colour Colour

Naming Naming (Vis)(Vis)

(Rec) (Rec) (Vis) (Vis)

Object Object

Viewing Viewing Colour Colour

Viewing Viewing

--Price & Friston (1997) Neuroimage

Example: Example:

Which neural structures support object recognition, Which neural structures support object recognition,

independent of task (naming vs viewing)?independent of task (naming vs viewing)?

ConjunctionsConjunctionsConjunctionsConjunctions

A1A1 A2A2

B2B2B1B1

Task (1/2)Task (1/2)

Stim

uli (

A/B

)

Col

ours

O

bjec

ts

Naming ViewingNaming Viewing

Price & Friston (1997) Neuroimage

Common object Common object recognition response (Recrecognition response (Rec))

B1 A1 B2 A2B1 A1 B2 A2

ConjunctionsConjunctionsConjunctionsConjunctions

•Test of global null hypothesis: Significant set of consistent effects

‣ “which voxels show effects of similar direction across contrasts?”

‣ does not correspond to logical AND

• Test of conjunction null hypothesis: Set of consistently significant effects

‣ “which voxels show for each contrast effects > threshold?”

‣ corresponds to logical AND

• Choice of test depends on hypothesis and congruence of contrasts; the global null test is more sensitive

Two flavours of inference about Two flavours of inference about conjunctionsconjunctions

Two flavours of inference about Two flavours of inference about conjunctionsconjunctions

Friston et al. (2005) Neuroimage

Nichols et al (2005) Neuroimage

• Categorical designsSubtraction - pure insertion, baseline problems

Conjunction - testing multiple hypotheses

• Parametric designsLinear - adaptation, cognitive dimensions

Nonlinear - polynomial expansion

Model-based - algorithmic definition

• Factorial designsCategorical - Main effects and Interactions

Parametric - Psychophysiological interactions

OverviewOverviewOverviewOverview

• Parametric designs approach the baseline problem by:- Varying an experimental parameter on a continuum, in multiple (n>2) steps...

- ... and relating blood-flow to this parameter

- No limit on the number steps (continuous scale)

Parametric designsParametric designsParametric designsParametric designs

• Such parameters can reflect many things:- Stimulus and task properties (e.g., color intensity, item difficulty, rated

attractiveness)- Experimental contexts (e.g., time since last presentation of same stimulus)- Participant performance (e.g., reaction time, degree of certainty)

• Flexible choice of tests for BOLD-parameter relations:- „Data-driven“ (e.g., neurometric functions for limited number of steps)- Linear- Nonlinear: Quadratic/cubic/etc.- Model-based

• Example: Varying word presentation rate

Parametric designs: Linear and nonlinear Parametric designs: Linear and nonlinear effectseffects

Parametric designs: Linear and nonlinear Parametric designs: Linear and nonlinear effectseffects

Rees et al (1997) Neuroimage

Right auditory cortex

SPM{F}

Left DLPFC

Linear contrast: Nonlinear contrast:• Advantage:

- Data-driven characterisation of BOLD-parameter relation

• Disadvantages:

- only possible for small number of parameter steps

- reduced statistical power (less df)

- contrasts often hard to define

- linear and non-linear components not properly separated

Model I: Separate regressorsModel I: Separate regressorsModel I: Separate regressorsModel I: Separate regressors

Polynomial expansion:f(x) ~ b0 + b1 x + b2 x2 + ...

…up to (N-1)th order for N levels

Model 2: Parametric modulationModel 2: Parametric modulationModel 2: Parametric modulationModel 2: Parametric modulation

Buechel et al (1998) Neuroimage

Polynomial expansion in SPMPolynomial expansion in SPMPolynomial expansion in SPMPolynomial expansion in SPM

Polynomial expansion in SPMPolynomial expansion in SPMPolynomial expansion in SPMPolynomial expansion in SPM

Mean correctionand

polynomial expansion and

orthogonalisation

- Separation of linear and non-linear effects on different regressors

- F-test on nonlinear regressor controls for linear and mean effects Buechel et al (1998) Neuroimage

Parametric Designs: Neurometric Parametric Designs: Neurometric functionsfunctions

Parametric Designs: Neurometric Parametric Designs: Neurometric functionsfunctions

– P0-P4: Variation of intensity of a laser stimulus applied to the right hand

(0, 300, 400, 500, 600 mJ)

– Neural coding of different aspects of tactile perception??

Stimulus awareness Stimulus intensity Pain intensityStimulus awareness Stimulus intensity Pain intensity

Buechel et al (2002) JNeurosci

Parametric Designs: Neurometric Parametric Designs: Neurometric functionsfunctions

Parametric Designs: Neurometric Parametric Designs: Neurometric functionsfunctions

Stimulus presenceStimulus presence

Pain intensityPain intensity

Stimulus intensityStimulus intensity

Buechel et al (2002) JNeurosci

Parametric Designs: Model-based Parametric Designs: Model-based regressorsregressors

Parametric Designs: Model-based Parametric Designs: Model-based regressorsregressors

• Parameters can be derived from computational model that specifies a neural mechanism, based on individual experimental context

• These “model-based” parameters can be included in the design matrix like any other parameter, and related to BOLD activity

O’Doherty et al (2003) Neuron

• Example: VS correlates with TD prediction error during appetitive conditioning

• Categorical designsSubtraction - pure insertion, baseline problems

Conjunction - testing multiple hypotheses

• Parametric designsLinear - adaptation, cognitive dimensions

Nonlinear - polynomial expansion

Model-based - algorithmic definition

• Factorial designsCategorical - Main effects and Interactions

Parametric - Psychophysiological interactions

OverviewOverviewOverviewOverview

Factorial designs: General principleFactorial designs: General principleFactorial designs: General principleFactorial designs: General principle

A1A1 A2A2

B2B2B1B1

Factor 1Factor 1

Fac

tor

2

......

......

............

• Combining n>2 factors (categorical and/or parametric)

• Fully balanced: Each factor step is paired with each step of the other factor, ideally with the same number of events

• Balanced factorial designs allow you to address the maximum number of questions with equal sensitivity

Factorial designs: Main effects and Factorial designs: Main effects and InteractionsInteractions

Factorial designs: Main effects and Factorial designs: Main effects and InteractionsInteractions

A1A1 A2A2

B2B2B1B1

Task (1/2)Task (1/2)

Stim

uli (

A/B

)

Col

ours

O

bjec

ts

Naming ViewingNaming Viewing

Factorial designs: Main effects and Factorial designs: Main effects and InteractionsInteractions

Factorial designs: Main effects and Factorial designs: Main effects and InteractionsInteractions

A1A1 A2A2

B2B2B1B1

Task (1/2)Task (1/2)

Stim

uli (

A/B

)

Col

ours

O

bjec

ts

Naming ViewingNaming Viewing

‣ failure of pure insertion

• Main effect (and conjunction) of task:

• (A1 + B1) – (A2 + B2)

Main effect (and conjunction) of stimuli:

• (A1 + A2) – (B1 + B2)

• Interaction of task and stimuli: • (A1 – B1) – (A2 –

B2)

Interactions and pure insertionInteractions and pure insertionInteractions and pure insertionInteractions and pure insertion

Interactions:

simple

and

cross-over

We can selectively inspect our data for one or the other by masking during inference

O CO C O CO C Nam Nam ViewView

O CO C O CO C Nam Nam ViewView

• Categorical designsSubtraction - pure insertion, baseline problems

Conjunction - testing multiple hypotheses

• Parametric designsLinear - adaptation, cognitive dimensions

Nonlinear - polynomial expansion

Model-based - algorithmic definition

• Factorial designsCategorical - Main effects and Interactions

Parametric - Psychophysiological interactions

OverviewOverviewOverviewOverview

ContextContextContextContext

sourcesourcesourcesource

targettargettargettarget

XXXX

Parametric factorial design in which one factor is a psychological context …

...and the other is a physiological source (activity extracted from a brain region of interest)

Psycho-physiological interactions (PPIs)Psycho-physiological interactions (PPIs)Psycho-physiological interactions (PPIs)Psycho-physiological interactions (PPIs)

PPIs: ExamplePPIs: ExamplePPIs: ExamplePPIs: Example

Changes in connectivity of V1 with attention to motion?

V1 Att V1xAtt

SPM{Z}SPM{Z}

attention

no attention

V1 activityV1 activity

V5

activ

ity

timetime

V1

activ

ity

Friston et al (1997) Neuroimage

ContextContextContextContext

sourcesourcesourcesource

targettargettargettarget

PPIs: Two possible interpretationsPPIs: Two possible interpretationsPPIs: Two possible interpretationsPPIs: Two possible interpretations

ContextContextContextContext

sourcesourcesourcesource

targettargettargettarget

Context-sensitive connectivity Modulation of

stimulus-related responses

Knowledge about neurobiological plausibility of both interpretations often necessary for interpreting PPI results

AttentionAttentionAttentionAttention

V1V1V1V1

V5V5V5V5

Knowledge about neurobiological plausibility of both interpretations often necessary for interpreting PPI results

AttentionAttentionAttentionAttention

V1V1V1V1

V5V5V5V5

Attention modulates V1- V5 connectivity V1 modulates impact of attention on V5

PPIs: Two possible interpretationsPPIs: Two possible interpretationsPPIs: Two possible interpretationsPPIs: Two possible interpretations

PPI: Regressor constructionPPI: Regressor constructionPPI: Regressor constructionPPI: Regressor construction

• PPI tested by a GLM with form:y = (V1xA).b1 + V1.b2 + A.b3 + e

• SPM provides routines that construct PPI regressors based on region timeseries and existing SPM.mat

• For interaction term (V1xA), we are interested in the interaction at the neural level, before convolution with HRF

• SPM thus deconvolves timeseries (V1) before producing interaction term (V1xA) and reconvolving with HRF